Project Summary: Recent years'CASP experiments have seen significant progress on computer-based protein tertiary structure prediction. The state-of-the-art algorithms, such as TASSER, can generate correct folds for 2/3 of single domain proteins without using homologous templates. Most of the models are of low resolution with a root mean square deviation (RMSD) 3-6 Angstroms, which may be useful for protein topology analysis and some level of biological function inference. However, such models are not suitable for reliable ligand screening. The proposed research seeks to develop new methodologies for the atomic-level high-resolution protein structure modeling. The specific aims are: (1) development of a new algorithm, atom-TASSER, for atomic-level protein structure modeling;(2) development of a two-level composite potential for both the low-resolution topology construction and the high-resolution refinement;(3) force field optimization in both structure and sequence phase spaces;(4) large-scale benchmarking of atom-TASSER algorithm. Overall, these studies are designed to extend the range of validity of the reduced modeling approaches, and provide significant improvements in the state-of-the-art of tertiary structure prediction. The goal is to push up the level of accuracy of computer-based modeling, especially for the distant/weakly homologous proteins, so that they can be of real use to new drug screening and biochemical functional inference.